ReadCtrl: Personalizing text generation with readability-controlled instruction learning

TL;DR

The results show that the ReadCtrl-Mistral-7B models significantly outperformed strong baseline models such as GPT-4 and Claude-3, with a win rate of 52.1%:35.7% against GPT-4 in human evaluations.

Abstract

Content generation conditioning on users's readability is an important application for personalization. In an era of large language models (LLMs), readability-controlled text generation based on LLMs has become increasingly important. This paper introduces a novel methodology called "Readability-Controlled Instruction Learning (ReadCtrl)," which aims to instruction-tune LLMs to tailor users' readability levels. Unlike the traditional methods, which primarily focused on categorical readability adjustments typically classified as high, medium, and low or expert and layperson levels with limited success, ReadCtrl introduces a dynamic framework that enables LLMs to generate content at various (near continuous level) complexity levels, thereby enhancing their versatility across different applications. Our results show that the ReadCtrl-Mistral-7B models significantly outperformed strong baseline models such as GPT-4 and Claude-3, with a win rate of 52.1%:35.7% against GPT-4 in human evaluations. Furthermore, Read-Ctrl has shown significant improvements in automatic evaluations, as evidenced by better readability metrics (e.g., FOG, FKGL) and generation quality metrics (e.g., BLEU, SARI, SummaC-Factuality, UniEval-Consistency and Coherence). These results underscore Read-Ctrl's effectiveness and tenacity in producing high-quality, contextually appropriate outputs that closely align with targeted readability levels, marking a significant advancement in personalized content generation using LLMs.

Figures (11)

• Figure 1: ReadCtrl instruction following ability. While current SOTA LLMs such as GPT and Claude (under the few-shot setting) show an upward trend in aligning their output with the target grade level, they fall significantly short of the 'perfect' adherence curve. Other weaker LLMs like Mistral-7b demonstrate little to no capacity to adjust to ReadCtrl instructions, as indicated by the flat line parallel to the x-axis. Notably, Mistral-ReadCtrl's performance closely approaches 'perfect', showcasing its advanced capability to tailor output to the specified readability level as set out by ReadCtrl instructions.
• Figure 2: Overview of ReadCtrl data construction.
• Figure 3: Win rate (%) for Mistral-ReadCtril vs GPT-4 (3 shots) using AI (Claude3 and GPT3.5) and Human evaluation.
• Figure 4: Screenshot of the human evaluation.
• Figure 5: Distribution of examples readability scores from instruction tuning datasets